Deep Learning Based Plant Diseases Monitoring and Detection System

Abstract

Farmers can identify plant illnesses by looking for spots on leaves and leaf yellowing, all of which occur based on previous observations. Yet, experts think that technology can help farmers with plant disease detection. Researchers developed a system that enables farmers to easily upload a picture of a leaf and obtain details about it, such as whether it is healthy or, in the case that it is not alternative remedies. For this investigation, researchers used CNNs, a subset of ANNs.4,500 leaf photos from four different classes made up the training set for our CNN model. Single-leafed plants with a uniform background and no additional noise make up the Plant Village dataset. But, the photos in the real scenario may differ. Hence, to improve the outcome in both the simulation and the real world, employed the YOLO technique to calculate ROI and then fed that ROI into the CNN model, which predicted our outcome. The CNN model contains a total of 10 layers: 2 sequential layers (for input preprocessing and augmentation), three convolutional layers, three pooling levels, one layer that is fully connected, and one output layer with a Softmax activation function. Three sections of our dataset were created: training (60%) validation (20%) and testing (20%). In this research, training the model throughout 60 epochs. The model is proposed, created, and trained with its validation accuracy and results show 96% and its test accuracy was 93%. Plant disease detection, deep learning, and ROI detection are the index terms.